"""Regresssion tests for urllib""" import urllib import httplib import unittest from test import test_support import os import mimetools import tempfile import StringIO import ftplib import threading import socket import time def hexescape(char): """Escape char as RFC 2396 specifies""" hex_repr = hex(ord(char))[2:].upper() if len(hex_repr) == 1: hex_repr = "0%s" % hex_repr return "%" + hex_repr class urlopen_FileTests(unittest.TestCase): """Test urlopen() opening a temporary file. Try to test as much functionality as possible so as to cut down on reliance on connecting to the Net for testing. """ def setUp(self): """Setup of a temp file to use for testing""" self.text = "test_urllib: %s\n" % self.__class__.__name__ FILE = open(test_support.TESTFN, 'wb') try: FILE.write(self.text) finally: FILE.close() self.pathname = test_support.TESTFN self.returned_obj = urllib.urlopen("file:%s" % self.pathname) def tearDown(self): """Shut down the open object""" self.returned_obj.close() os.remove(test_support.TESTFN) def test_interface(self): # Make sure object returned by urlopen() has the specified methods for attr in ("read", "readline", "readlines", "fileno", "close", "info", "geturl", "__iter__"): self.assert_(hasattr(self.returned_obj, attr), "object returned by urlopen() lacks %s attribute" % attr) def test_read(self): self.assertEqual(self.text, self.returned_obj.read()) def test_readline(self): self.assertEqual(self.text, self.returned_obj.readline()) self.assertEqual('', self.returned_obj.readline(), "calling readline() after exhausting the file did not" " return an empty string") def test_readlines(self): lines_list = self.returned_obj.readlines() self.assertEqual(len(lines_list), 1, "readlines() returned the wrong number of lines") self.assertEqual(lines_list[0], self.text, "readlines() returned improper text") def test_fileno(self): file_num = self.returned_obj.fileno() self.assert_(isinstance(file_num, int), "fileno() did not return an int") self.assertEqual(os.read(file_num, len(self.text)), self.text, "Reading on the file descriptor returned by fileno() " "did not return the expected text") def test_close(self): # Test close() by calling it hear and then having it be called again # by the tearDown() method for the test self.returned_obj.close() def test_info(self): self.assert_(isinstance(self.returned_obj.info(), mimetools.Message)) def test_geturl(self): self.assertEqual(self.returned_obj.geturl(), self.pathname) def test_iter(self): # Test iterator # Don't need to count number of iterations since test would fail the # instant it returned anything beyond the first line from the # comparison for line in self.returned_obj.__iter__(): self.assertEqual(line, self.text) class urlopen_HttpTests(unittest.TestCase): """Test urlopen() opening a fake http connection.""" def fakehttp(self, fakedata): class FakeSocket(StringIO.StringIO): def sendall(self, str): pass def makefile(self, mode, name): return self def read(self, amt=None): if self.closed: return '' return StringIO.StringIO.read(self, amt) def readline(self, length=None): if self.closed: return '' return StringIO.StringIO.readline(self, length) class FakeHTTPConnection(httplib.HTTPConnection): def connect(self): self.sock = FakeSocket(fakedata) assert httplib.HTTP._connection_class == httplib.HTTPConnection httplib.HTTP._connection_class = FakeHTTPConnection def unfakehttp(self): httplib.HTTP._connection_class = httplib.HTTPConnection def test_read(self): self.fakehttp('Hello!') try: fp = urllib.urlopen("http://python.org/") self.assertEqual(fp.readline(), 'Hello!') self.assertEqual(fp.readline(), '') finally: self.unfakehttp() def test_empty_socket(self): """urlopen() raises IOError if the underlying socket does not send any data. (#1680230) """ self.fakehttp('') try: self.assertRaises(IOError, urllib.urlopen, 'http://something') finally: self.unfakehttp() class urlretrieve_FileTests(unittest.TestCase): """Test urllib.urlretrieve() on local files""" def setUp(self): # Create a list of temporary files. Each item in the list is a file # name (absolute path or relative to the current working directory). # All files in this list will be deleted in the tearDown method. Note, # this only helps to makes sure temporary files get deleted, but it # does nothing about trying to close files that may still be open. It # is the responsibility of the developer to properly close files even # when exceptional conditions occur. self.tempFiles = [] # Create a temporary file. self.registerFileForCleanUp(test_support.TESTFN) self.text = 'testing urllib.urlretrieve' try: FILE = open(test_support.TESTFN, 'wb') FILE.write(self.text) FILE.close() finally: try: FILE.close() except: pass def tearDown(self): # Delete the temporary files. for each in self.tempFiles: try: os.remove(each) except: pass def constructLocalFileUrl(self, filePath): return "file://%s" % urllib.pathname2url(os.path.abspath(filePath)) def createNewTempFile(self, data=""): """Creates a new temporary file containing the specified data, registers the file for deletion during the test fixture tear down, and returns the absolute path of the file.""" newFd, newFilePath = tempfile.mkstemp() try: self.registerFileForCleanUp(newFilePath) newFile = os.fdopen(newFd, "wb") newFile.write(data) newFile.close() finally: try: newFile.close() except: pass return newFilePath def registerFileForCleanUp(self, fileName): self.tempFiles.append(fileName) def test_basic(self): # Make sure that a local file just gets its own location returned and # a headers value is returned. result = urllib.urlretrieve("file:%s" % test_support.TESTFN) self.assertEqual(result[0], test_support.TESTFN) self.assert_(isinstance(result[1], mimetools.Message), "did not get a mimetools.Message instance as second " "returned value") def test_copy(self): # Test that setting the filename argument works. second_temp = "%s.2" % test_support.TESTFN self.registerFileForCleanUp(second_temp) result = urllib.urlretrieve(self.constructLocalFileUrl( test_support.TESTFN), second_temp) self.assertEqual(second_temp, result[0]) self.assert_(os.path.exists(second_temp), "copy of the file was not " "made") FILE = open(second_temp, 'rb') try: text = FILE.read() FILE.close() finally: try: FILE.close() except: pass self.assertEqual(self.text, text) def test_reporthook(self): # Make sure that the reporthook works. def hooktester(count, block_size, total_size, count_holder=[0]): self.assert_(isinstance(count, int)) self.assert_(isinstance(block_size, int)) self.assert_(isinstance(total_size, int)) self.assertEqual(count, count_holder[0]) count_holder[0] = count_holder[0] + 1 second_temp = "%s.2" % test_support.TESTFN self.registerFileForCleanUp(second_temp) urllib.urlretrieve(self.constructLocalFileUrl(test_support.TESTFN), second_temp, hooktester) def test_reporthook_0_bytes(self): # Test on zero length file. Should call reporthook only 1 time. report = [] def hooktester(count, block_size, total_size, _report=report): _report.append((count, block_size, total_size)) srcFileName = self.createNewTempFile() urllib.urlretrieve(self.constructLocalFileUrl(srcFileName), test_support.TESTFN, hooktester) self.assertEqual(len(report), 1) self.assertEqual(report[0][2], 0) def test_reporthook_5_bytes(self): # Test on 5 byte file. Should call reporthook only 2 times (once when # the "network connection" is established and once when the block is # read). Since the block size is 8192 bytes, only one block read is # required to read the entire file. report = [] def hooktester(count, block_size, total_size, _report=report): _report.append((count, block_size, total_size)) srcFileName = self.createNewTempFile("x" * 5) urllib.urlretrieve(self.constructLocalFileUrl(srcFileName), test_support.TESTFN, hooktester) self.assertEqual(len(report), 2) self.assertEqual(report[0][1], 8192) self.assertEqual(report[0][2], 5) def test_reporthook_8193_bytes(self): # Test on 8193 byte file. Should call reporthook only 3 times (once # when the "network connection" is established, once for the next 8192 # bytes, and once for the last byte). report = [] def hooktester(count, block_size, total_size, _report=report): _report.append((count, block_size, total_size)) srcFileName = self.createNewTempFile("x" * 8193) urllib.urlretrieve(self.constructLocalFileUrl(srcFileName), test_support.TESTFN, hooktester) self.assertEqual(len(report), 3) self.assertEqual(report[0][1], 8192) self.assertEqual(report[0][2], 8193) class QuotingTests(unittest.TestCase): """Tests for urllib.quote() and urllib.quote_plus() According to RFC 2396 ("Uniform Resource Identifiers), to escape a character you write it as '%' + <2 character US-ASCII hex value>. The Python code of ``'%' + hex(ord())[2:]`` escapes a character properly. Case does not matter on the hex letters. The various character sets specified are: Reserved characters : ";/?:@&=+$," Have special meaning in URIs and must be escaped if not being used for their special meaning Data characters : letters, digits, and "-_.!~*'()" Unreserved and do not need to be escaped; can be, though, if desired Control characters : 0x00 - 0x1F, 0x7F Have no use in URIs so must be escaped space : 0x20 Must be escaped Delimiters : '<>#%"' Must be escaped Unwise : "{}|\^[]`" Must be escaped """ def test_never_quote(self): # Make sure quote() does not quote letters, digits, and "_,.-" do_not_quote = '' .join(["ABCDEFGHIJKLMNOPQRSTUVWXYZ", "abcdefghijklmnopqrstuvwxyz", "0123456789", "_.-"]) result = urllib.quote(do_not_quote) self.assertEqual(do_not_quote, result, "using quote(): %s != %s" % (do_not_quote, result)) result = urllib.quote_plus(do_not_quote) self.assertEqual(do_not_quote, result, "using quote_plus(): %s != %s" % (do_not_quote, result)) def test_default_safe(self): # Test '/' is default value for 'safe' parameter self.assertEqual(urllib.quote.__defaults__[0], '/') def test_safe(self): # Test setting 'safe' parameter does what it should do quote_by_default = "<>" result = urllib.quote(quote_by_default, safe=quote_by_default) self.assertEqual(quote_by_default, result, "using quote(): %s != %s" % (quote_by_default, result)) result = urllib.quote_plus(quote_by_default, safe=quote_by_default) self.assertEqual(quote_by_default, result, "using quote_plus(): %s != %s" % (quote_by_default, result)) def test_default_quoting(self): # Make sure all characters that should be quoted are by default sans # space (separate test for that). should_quote = [chr(num) for num in range(32)] # For 0x00 - 0x1F should_quote.append('<>#%"{}|\^[]`') should_quote.append(chr(127)) # For 0x7F should_quote = ''.join(should_quote) for char in should_quote: result = urllib.quote(char) self.assertEqual(hexescape(char), result, "using quote(): %s should be escaped to %s, not %s" % (char, hexescape(char), result)) result = urllib.quote_plus(char) self.assertEqual(hexescape(char), result, "using quote_plus(): " "%s should be escapes to %s, not %s" % (char, hexescape(char), result)) del should_quote partial_quote = "ab[]cd" expected = "ab%5B%5Dcd" result = urllib.quote(partial_quote) self.assertEqual(expected, result, "using quote(): %s != %s" % (expected, result)) self.assertEqual(expected, result, "using quote_plus(): %s != %s" % (expected, result)) def test_quoting_space(self): # Make sure quote() and quote_plus() handle spaces as specified in # their unique way result = urllib.quote(' ') self.assertEqual(result, hexescape(' '), "using quote(): %s != %s" % (result, hexescape(' '))) result = urllib.quote_plus(' ') self.assertEqual(result, '+', "using quote_plus(): %s != +" % result) given = "a b cd e f" expect = given.replace(' ', hexescape(' ')) result = urllib.quote(given) self.assertEqual(expect, result, "using quote(): %s != %s" % (expect, result)) expect = given.replace(' ', '+') result = urllib.quote_plus(given) self.assertEqual(expect, result, "using quote_plus(): %s != %s" % (expect, result)) def test_quoting_plus(self): self.assertEqual(urllib.quote_plus('alpha+beta gamma'), 'alpha%2Bbeta+gamma') self.assertEqual(urllib.quote_plus('alpha+beta gamma', '+'), 'alpha+beta+gamma') class UnquotingTests(unittest.TestCase): """Tests for unquote() and unquote_plus() See the doc string for quoting_Tests for details on quoting and such. """ def test_unquoting(self): # Make sure unquoting of all ASCII values works escape_list = [] for num in range(128): given = hexescape(chr(num)) expect = chr(num) result = urllib.unquote(given) self.assertEqual(expect, result, "using unquote(): %s != %s" % (expect, result)) result = urllib.unquote_plus(given) self.assertEqual(expect, result, "using unquote_plus(): %s != %s" % (expect, result)) escape_list.append(given) escape_string = ''.join(escape_list) del escape_list result = urllib.unquote(escape_string) self.assertEqual(result.count('%'), 1, "using quote(): not all characters escaped; %s" % result) result = urllib.unquote(escape_string) self.assertEqual(result.count('%'), 1, "using unquote(): not all characters escaped: " "%s" % result) def test_unquoting_parts(self): # Make sure unquoting works when have non-quoted characters # interspersed given = 'ab%sd' % hexescape('c') expect = "abcd" result = urllib.unquote(given) self.assertEqual(expect, result, "using quote(): %s != %s" % (expect, result)) result = urllib.unquote_plus(given) self.assertEqual(expect, result, "using unquote_plus(): %s != %s" % (expect, result)) def test_unquoting_plus(self): # Test difference between unquote() and unquote_plus() given = "are+there+spaces..." expect = given result = urllib.unquote(given) self.assertEqual(expect, result, "using unquote(): %s != %s" % (expect, result)) expect = given.replace('+', ' ') result = urllib.unquote_plus(given) self.assertEqual(expect, result, "using unquote_plus(): %s != %s" % (expect, result)) def test_unquote_with_unicode(self): r = urllib.unquote('br%C3%BCckner_sapporo_20050930.doc') self.assertEqual(r, 'br\xc3\xbcckner_sapporo_20050930.doc') class urlencode_Tests(unittest.TestCase): """Tests for urlencode()""" def help_inputtype(self, given, test_type): """Helper method for testing different input types. 'given' must lead to only the pairs: * 1st, 1 * 2nd, 2 * 3rd, 3 Test cannot assume anything about order. Docs make no guarantee and have possible dictionary input. """ expect_somewhere = ["1st=1", "2nd=2", "3rd=3"] result = urllib.urlencode(given) for expected in expect_somewhere: self.assert_(expected in result, "testing %s: %s not found in %s" % (test_type, expected, result)) self.assertEqual(result.count('&'), 2, "testing %s: expected 2 '&'s; got %s" % (test_type, result.count('&'))) amp_location = result.index('&') on_amp_left = result[amp_location - 1] on_amp_right = result[amp_location + 1] self.assert_(on_amp_left.isdigit() and on_amp_right.isdigit(), "testing %s: '&' not located in proper place in %s" % (test_type, result)) self.assertEqual(len(result), (5 * 3) + 2, #5 chars per thing and amps "testing %s: " "unexpected number of characters: %s != %s" % (test_type, len(result), (5 * 3) + 2)) def test_using_mapping(self): # Test passing in a mapping object as an argument. self.help_inputtype({"1st":'1', "2nd":'2', "3rd":'3'}, "using dict as input type") def test_using_sequence(self): # Test passing in a sequence of two-item sequences as an argument. self.help_inputtype([('1st', '1'), ('2nd', '2'), ('3rd', '3')], "using sequence of two-item tuples as input") def test_quoting(self): # Make sure keys and values are quoted using quote_plus() given = {"&":"="} expect = "%s=%s" % (hexescape('&'), hexescape('=')) result = urllib.urlencode(given) self.assertEqual(expect, result) given = {"key name":"A bunch of pluses"} expect = "key+name=A+bunch+of+pluses" result = urllib.urlencode(given) self.assertEqual(expect, result) def test_doseq(self): # Test that passing True for 'doseq' parameter works correctly given = {'sequence':['1', '2', '3']} expect = "sequence=%s" % urllib.quote_plus(str(['1', '2', '3'])) result = urllib.urlencode(given) self.assertEqual(expect, result) result = urllib.urlencode(given, True) for value in given["sequence"]: expect = "sequence=%s" % value self.assert_(expect in result, "%s not found in %s" % (expect, result)) self.assertEqual(result.count('&'), 2, "Expected 2 '&'s, got %s" % result.count('&')) class Pathname_Tests(unittest.TestCase): """Test pathname2url() and url2pathname()""" def test_basic(self): # Make sure simple tests pass expected_path = os.path.join("parts", "of", "a", "path") expected_url = "parts/of/a/path" result = urllib.pathname2url(expected_path) self.assertEqual(expected_url, result, "pathname2url() failed; %s != %s" % (result, expected_url)) result = urllib.url2pathname(expected_url) self.assertEqual(expected_path, result, "url2pathame() failed; %s != %s" % (result, expected_path)) def test_quoting(self): # Test automatic quoting and unquoting works for pathnam2url() and # url2pathname() respectively given = os.path.join("needs", "quot=ing", "here") expect = "needs/%s/here" % urllib.quote("quot=ing") result = urllib.pathname2url(given) self.assertEqual(expect, result, "pathname2url() failed; %s != %s" % (expect, result)) expect = given result = urllib.url2pathname(result) self.assertEqual(expect, result, "url2pathname() failed; %s != %s" % (expect, result)) given = os.path.join("make sure", "using_quote") expect = "%s/using_quote" % urllib.quote("make sure") result = urllib.pathname2url(given) self.assertEqual(expect, result, "pathname2url() failed; %s != %s" % (expect, result)) given = "make+sure/using_unquote" expect = os.path.join("make+sure", "using_unquote") result = urllib.url2pathname(given) self.assertEqual(expect, result, "url2pathname() failed; %s != %s" % (expect, result)) # Just commented them out. # Can't really tell why keep failing in windows and sparc. # Everywhere else they work ok, but on those machines, someteimes # fail in one of the tests, sometimes in other. I have a linux, and # the tests go ok. # If anybody has one of the problematic enviroments, please help! # . Facundo # # def server(evt): # serv = socket.socket(socket.AF_INET, socket.SOCK_STREAM) # serv.settimeout(3) # serv.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) # serv.bind(("", 9093)) # serv.listen(5) # try: # conn, addr = serv.accept() # conn.send("1 Hola mundo\n") # cantdata = 0 # while cantdata < 13: # data = conn.recv(13-cantdata) # cantdata += len(data) # time.sleep(.3) # conn.send("2 No more lines\n") # conn.close() # except socket.timeout: # pass # finally: # serv.close() # evt.set() # # class FTPWrapperTests(unittest.TestCase): # # def setUp(self): # ftplib.FTP.port = 9093 # self.evt = threading.Event() # threading.Thread(target=server, args=(self.evt,)).start() # time.sleep(.1) # # def tearDown(self): # self.evt.wait() # # def testBasic(self): # # connects # ftp = urllib.ftpwrapper("myuser", "mypass", "localhost", 9093, []) # ftp.ftp.sock.close() # # def testTimeoutDefault(self): # # default # ftp = urllib.ftpwrapper("myuser", "mypass", "localhost", 9093, []) # self.assertTrue(ftp.ftp.sock.gettimeout() is None) # ftp.ftp.sock.close() # # def testTimeoutValue(self): # # a value # ftp = urllib.ftpwrapper("myuser", "mypass", "localhost", 9093, [], timeout=30) # self.assertEqual(ftp.ftp.sock.gettimeout(), 30) # ftp.ftp.sock.close() # # def testTimeoutNone(self): # # None, having other default # previous = socket.getdefaulttimeout() # socket.setdefaulttimeout(30) # try: # ftp = urllib.ftpwrapper("myuser", "mypass", "localhost", 9093, []) # finally: # socket.setdefaulttimeout(previous) # self.assertEqual(ftp.ftp.sock.gettimeout(), 30) # ftp.ftp.close() # def test_main(): test_support.run_unittest( urlopen_FileTests, urlopen_HttpTests, urlretrieve_FileTests, QuotingTests, UnquotingTests, urlencode_Tests, Pathname_Tests, #FTPWrapperTests, ) if __name__ == '__main__': test_main()